A multiple imputation joint model to estimate COVID-19 correlates of protection, antibody decay and vaccine efficacy waning

Thesis

Estimating correlates of protection – the relationship between an immune marker and risk of infection – is a key task of vaccine research. Standard models to estimate a correlate of protection do not account for the decay of antibody levels over time. We demonstrate that when differing background infection rates cause the timing of cases to vary between trials, the estimated correlate of protection curve will not be consistent. This inconsistency is caused by antibody decay between immune marker measurement and infection. Therefore, methods are required which estimate the relationship between antibody levels at the time of exposure and subsequent protection, sometimes known as “exposure-proximal” correlates of protection. We develop a joint model of waning antibody levels and subsequent COVID-19 infection. We apply the model to data from a COVID-19 vaccine trial, to estimate correlates of protection at exposure, antibody decay, and waning vaccine efficacy. The model uses a two-stage multiple imputation approach, and pools the results via an approximate Bayesian pooling scheme. We extend our multiple imputation pooling method to other scenarios, and explore how our approach compares to Rubin's pooling rules mathematically. We compare Rubin's rules and Bayesian inference with our approach in a simulation study, and in real data examples. The proposed approximate Bayesian pooling method is able to account for skewness in the missing-data posterior, whereas Rubin's rules assumes the distribution is symmetric.

Authors: Phillips, DJ

• DOI: 10.5287/ora-bpvpyb92x
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: antibody decay; joint modelling; SARS-CoV-2; survival analysis; time-to-event data; exposure-proximal correlates of protection; correlates of protection; Bayesian inference; longitudinal data; multiple imputation; COVID-19 vaccines; missing data; vaccine efficacy waning
• Subjects: Statistics; COVID-19 vaccines; Missing data; Bayesian inference; Survival analysis